Exploring the Dual Nature of AI in Drug Discovery and Design
Written on
Chapter 1: The Promise of Machine Learning in Pharmaceuticals
Machine learning has emerged as a transformative force across various sectors, particularly in life sciences. As we delve deeper into the molecular landscape, we uncover two critical insights: 1) the sheer number of molecules involved in biological processes and 2) the diverse functionalities these molecules exhibit.
The combination of these factors results in an overwhelming amount of data.
“Knock, knock.”
“Who’s there?”
“AI.”
“AI who?”
“AI capable of sifting through vast molecular databases to detect patterns beyond human perception.”
(It is important to note that while machine learning is commonly associated with AI, they are not synonymous. Machine learning is a predominant technique for developing AI, but it is not the sole method. The terms are often used interchangeably.)
In prior discussions, we've explored how this AI capability can aid in combating diseases. We specifically examined two areas: drug repurposing and the design of new drugs.
Drug repurposing refers to the practice of employing existing medications for different conditions than those for which they were initially created. Research has utilized this approach to identify potential treatments for Alzheimer's disease and pharmaceuticals with promise for extending life. Without AI to guide us, these previously overlooked applications of existing drugs might have remained undiscovered. One significant advantage of drug repurposing is that data on safety and side effects is typically already available. However, it is crucial to remember that AI can make errors, necessitating thorough testing of the identified drugs for the conditions the AI suggested.
New drug design is relatively straightforward. AI and machine learning can accelerate this process. Large collections of compounds await exploration, and preliminary tests can be standardized and automated with relative ease (potentially managed by AI-driven robotics). When combined with knowledge of protein targets, this can significantly hasten the identification or creation of compounds with promising therapeutic effects for further research.
Section 1.1: The Dual-Use Dilemma
Yet, like many groundbreaking technologies, we must consider the dual-use dilemma associated with AI-driven drug design and repurposing. These systems can be harnessed for beneficial purposes, but they also hold the potential for misuse.
A recent study highlights this concern.
After being invited to discuss the potential misapplications of AI in molecular discovery, researchers employed their molecular prediction system—which typically aims to find therapeutic inhibitors for human diseases. Instead of seeking new medications, they shifted focus to identifying toxic substances.
Typically, their virtual molecule generator discourages toxicity while rewarding activity on therapeutic targets. This time, the researchers reversed their approach. Following some adjustments to the AI's data processing:
In less than six hours on their internal server, the model produced 40,000 molecules that met their specified criteria. Alarmingly, the AI not only created VX but also numerous known chemical warfare agents that were later confirmed through visual comparisons with structures in public chemical databases. Several new molecules were also generated, predicted to be more toxic than established chemical warfare agents based on their anticipated LD50 values.
LD50 refers to the median lethal dose, the amount needed to lethally affect half of a tested population within a designated timeframe.
This scenario raises serious concerns: the foundational generative software bears similarities to readily available open-source programs.
The authors of the study argue that this should act as a wake-up call, urging a serious dialogue regarding the dual-use potential of AI in drug discovery.
For us, the implications of repurposing machine learning have become a pressing issue. We must now confront the question: what are the broader implications of these advancements?
Section 1.2: The Call for Awareness
As always, thank you for your readership. You may be interested in my newsletter, Thinking Ahead, where I discuss broader themes in science, philosophy, technology, and occasionally venture into fiction. It’s free to subscribe, and I assure you, your inbox will remain spam-free.
